Claude Code - Everything Configurable That the Docs Don't Tell You

Based on @anthropic-ai/[email protected], this covers detailed configuration and control methods for Hooks, Skills, and Agents that are not specified in the official documentation.

Real-time Behavior Control via Hooks Hooks go beyond simply deciding whether to execute, allowing real-time modification of Claude Code's behavior through event-specific JSON fields in stdout.

• PreToolUse: You can modify commands before tool execution using updatedInput (e.g., changing git push to --dry-run), or enforce permissions using permissionDecision.
• SessionStart: You can watch specific files with watchPaths, or inject context that persists throughout the session via additionalContext.
• PostToolUse: You can use updatedMCPToolOutput to modify the MCP tool response that Claude receives.

Settings and Configuration System

• Settings Files: Personal settings are located at ~/.claude/settings.json, and project settings at .claude/settings.json, with project settings shareable with the team via Git.
• Skills & Agents: Hidden frontmatter allows fine-grained control over model type, scope, persistent memory, MCP dependencies, and more.

By leveraging these features, Claude Code can be configured as a learning-based development environment optimized for project characteristics, going beyond a simple assistant.